Why Probabilistic Forecasting Is Better for Inventory Optimization

Last month we published a primer on probabilistic forecasting, an alternative to deterministic or ‘single number’ forecasting.

In this post we’ll explore how probabilistic forecasting not only creates better forecasts, but also improves inventory optimization by improving the stock levels at store and warehouse locations in your distribution network.

Inventory Optimization: Making the Most of Your Probabilistic Forecast

To understand the relationship between probabilistic forecasting and inventory, view your supply chain as a dynamic system, subject to uncertainty and unpredictable change.

Spreadsheets and legacy suites like SAP APO produce top-down aggregated forecasts using a deterministic approach.

While easier to comprehend, in this environment they produce chronically poor forecasting outcomes.

However probabilistic forecasting doesn’t just create an average forecast; it identifies a range of outcomes and the probability of each of those outcomes occurring.

Inventory optimization software can then use this information to better identify the optimal inventory targets.

Getting into the Details

Here is a simplistic example.

A single number forecasting system might look at a sales history for a specific SKU of 12 tires per year and identify average demand as one tire per month.

Therefore it may propose keeping one tire in stock. Because this forecast does not address customers replacing all four tires at once, it would continually propose poor inventory stocking levels.

For inventory purposes, you really want to know the probability of each order quantity – for one tire, two tires, three tires, four tires, etc. which then provides better information from which to decide how much inventory to stock.

Probabilistic forecasting provides exactly that information, identifying the order patterns (e.g., order size, order frequency) that inventory can use to service demand.

Most real life examples are far more complex than this simple illustration, but the basic concept remains the same.

Understanding the demand details yields a clearer understanding of the forecast than just averages.

Industry Opinion and Real-World Examples

Nucleus Research says that inventory optimization is best understood as “a form of predictive analytics.”

They say that best-in-class probabilistic (also referred to as “stochastic”) planning systems do the most accurate job of predicting the amount and type of stock to carry at the item-location level.

A deterministic approach is particularly inappropriate for forecasting for items with intermittent demand such as specialty goods or spare parts.

That’s because demand for “long tail” items is intermittent and doesn’t conform to predictions of “average” demand or normal distributions.

So demand details make all the difference.

Real life examples of companies using probability-based planning systems abound.

Consumer giant P&G recently moved to a probabilistic forecasting tool to plan its complex Global Distributor Markets (GDM) supply chain that serves emerging and challenger markets.

Global prescription lens manufacturer Shamir Optical’s use of probabilistic forecasting illustrates how probability-based systems enable companies to be more “service-driven.”

Rather than a “one size fits all” inventory policy, by leveraging the demand patterns described above, Shamir used what’s known as “mix optimization” to create a blend of different service level targets for each individual SKU in each location.

For example, instead of assigning all SKUs in a class a 98% service level, a global 98% target is achieved by optimally setting individual SKUs service levels at 95%, 97%, 99.5%, etc., achieving the same overall objective with far less inventory expense.

The firm reduced inventory levels by more than 25 percent overall while consistently achieving service levels exceeding 99 percent.

Probabilistic forecasting yields not just better forecasts, but leads to a host of inventory planning and optimization benefits as well, such better addressing specific demand patterns, long tail demand, complex supply chains, and achieving aggressive service level targets.

It’s clearly a winning approach to demand forecasting.